2 * Copyright (c) 2003-2007 Tim Kientzle
3 * Copyright (c) 2010-2012 Michihiro NAKAJIMA
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR(S) ``AS IS'' AND ANY EXPRESS OR
16 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
18 * IN NO EVENT SHALL THE AUTHOR(S) BE LIABLE FOR ANY DIRECT, INDIRECT,
19 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
20 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
21 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
22 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
23 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
24 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27 #include "archive_platform.h"
28 __FBSDID("$FreeBSD: head/lib/libarchive/archive_write_set_format_pax.c 201162 2009-12-29 05:47:46Z kientzle $");
41 #include "archive_entry.h"
42 #include "archive_entry_locale.h"
43 #include "archive_private.h"
44 #include "archive_write_private.h"
47 struct sparse_block
*next
;
54 uint64_t entry_bytes_remaining
;
55 uint64_t entry_padding
;
56 struct archive_string l_url_encoded_name
;
57 struct archive_string pax_header
;
58 struct archive_string sparse_map
;
59 size_t sparse_map_padding
;
60 struct sparse_block
*sparse_list
;
61 struct sparse_block
*sparse_tail
;
62 struct archive_string_conv
*sconv_utf8
;
66 static void add_pax_attr(struct archive_string
*, const char *key
,
68 static void add_pax_attr_int(struct archive_string
*,
69 const char *key
, int64_t value
);
70 static void add_pax_attr_time(struct archive_string
*,
71 const char *key
, int64_t sec
,
73 static ssize_t
archive_write_pax_data(struct archive_write
*,
74 const void *, size_t);
75 static int archive_write_pax_close(struct archive_write
*);
76 static int archive_write_pax_free(struct archive_write
*);
77 static int archive_write_pax_finish_entry(struct archive_write
*);
78 static int archive_write_pax_header(struct archive_write
*,
79 struct archive_entry
*);
80 static int archive_write_pax_options(struct archive_write
*,
81 const char *, const char *);
82 static char *base64_encode(const char *src
, size_t len
);
83 static char *build_gnu_sparse_name(char *dest
, const char *src
);
84 static char *build_pax_attribute_name(char *dest
, const char *src
);
85 static char *build_ustar_entry_name(char *dest
, const char *src
,
86 size_t src_length
, const char *insert
);
87 static char *format_int(char *dest
, int64_t);
88 static int has_non_ASCII(const char *);
89 static void sparse_list_clear(struct pax
*);
90 static int sparse_list_add(struct pax
*, int64_t, int64_t);
91 static char *url_encode(const char *in
);
94 * Set output format to 'restricted pax' format.
96 * This is the same as normal 'pax', but tries to suppress
97 * the pax header whenever possible. This is the default for
98 * bsdtar, for instance.
101 archive_write_set_format_pax_restricted(struct archive
*_a
)
103 struct archive_write
*a
= (struct archive_write
*)_a
;
106 archive_check_magic(_a
, ARCHIVE_WRITE_MAGIC
,
107 ARCHIVE_STATE_NEW
, "archive_write_set_format_pax_restricted");
109 r
= archive_write_set_format_pax(&a
->archive
);
110 a
->archive
.archive_format
= ARCHIVE_FORMAT_TAR_PAX_RESTRICTED
;
111 a
->archive
.archive_format_name
= "restricted POSIX pax interchange";
116 * Set output format to 'pax' format.
119 archive_write_set_format_pax(struct archive
*_a
)
121 struct archive_write
*a
= (struct archive_write
*)_a
;
124 archive_check_magic(_a
, ARCHIVE_WRITE_MAGIC
,
125 ARCHIVE_STATE_NEW
, "archive_write_set_format_pax");
127 if (a
->format_free
!= NULL
)
130 pax
= (struct pax
*)malloc(sizeof(*pax
));
132 archive_set_error(&a
->archive
, ENOMEM
,
133 "Can't allocate pax data");
134 return (ARCHIVE_FATAL
);
136 memset(pax
, 0, sizeof(*pax
));
137 a
->format_data
= pax
;
138 a
->format_name
= "pax";
139 a
->format_options
= archive_write_pax_options
;
140 a
->format_write_header
= archive_write_pax_header
;
141 a
->format_write_data
= archive_write_pax_data
;
142 a
->format_close
= archive_write_pax_close
;
143 a
->format_free
= archive_write_pax_free
;
144 a
->format_finish_entry
= archive_write_pax_finish_entry
;
145 a
->archive
.archive_format
= ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE
;
146 a
->archive
.archive_format_name
= "POSIX pax interchange";
151 archive_write_pax_options(struct archive_write
*a
, const char *key
,
154 struct pax
*pax
= (struct pax
*)a
->format_data
;
155 int ret
= ARCHIVE_FAILED
;
157 if (strcmp(key
, "hdrcharset") == 0) {
159 * The character-set we can use are defined in
160 * IEEE Std 1003.1-2001
162 if (val
== NULL
|| val
[0] == 0)
163 archive_set_error(&a
->archive
, ARCHIVE_ERRNO_MISC
,
164 "pax: hdrcharset option needs a character-set name");
165 else if (strcmp(val
, "BINARY") == 0 ||
166 strcmp(val
, "binary") == 0) {
168 * Specify binary mode. We will not convert
169 * filenames, uname and gname to any charsets.
173 } else if (strcmp(val
, "UTF-8") == 0) {
175 * Specify UTF-8 character-set to be used for
176 * filenames. This is almost the test that
177 * running platform supports the string conversion.
178 * Especially libarchive_test needs this trick for
181 pax
->sconv_utf8
= archive_string_conversion_to_charset(
182 &(a
->archive
), "UTF-8", 0);
183 if (pax
->sconv_utf8
== NULL
)
188 archive_set_error(&a
->archive
, ARCHIVE_ERRNO_MISC
,
189 "pax: invalid charset name");
193 /* Note: The "warn" return is just to inform the options
194 * supervisor that we didn't handle it. It will generate
195 * a suitable error if no one used this option. */
196 return (ARCHIVE_WARN
);
200 * Note: This code assumes that 'nanos' has the same sign as 'sec',
201 * which implies that sec=-1, nanos=200000000 represents -1.2 seconds
202 * and not -0.8 seconds. This is a pretty pedantic point, as we're
203 * unlikely to encounter many real files created before Jan 1, 1970,
204 * much less ones with timestamps recorded to sub-second resolution.
207 add_pax_attr_time(struct archive_string
*as
, const char *key
,
208 int64_t sec
, unsigned long nanos
)
213 * Note that each byte contributes fewer than 3 base-10
214 * digits, so this will always be big enough.
216 char tmp
[1 + 3*sizeof(sec
) + 1 + 3*sizeof(nanos
)];
218 tmp
[sizeof(tmp
) - 1] = 0;
219 t
= tmp
+ sizeof(tmp
) - 1;
221 /* Skip trailing zeros in the fractional part. */
222 for (digit
= 0, i
= 10; i
> 0 && digit
== 0; i
--) {
227 /* Only format the fraction if it's non-zero. */
230 *--t
= "0123456789"[digit
];
237 t
= format_int(t
, sec
);
239 add_pax_attr(as
, key
, t
);
243 format_int(char *t
, int64_t i
)
248 ui
= (i
== INT64_MIN
) ? (uint64_t)(INT64_MAX
) + 1 : (uint64_t)(-i
);
253 *--t
= "0123456789"[ui
% 10];
261 add_pax_attr_int(struct archive_string
*as
, const char *key
, int64_t value
)
263 char tmp
[1 + 3 * sizeof(value
)];
265 tmp
[sizeof(tmp
) - 1] = 0;
266 add_pax_attr(as
, key
, format_int(tmp
+ sizeof(tmp
) - 1, value
));
270 * Add a key/value attribute to the pax header. This function handles
271 * the length field and various other syntactic requirements.
274 add_pax_attr(struct archive_string
*as
, const char *key
, const char *value
)
276 int digits
, i
, len
, next_ten
;
277 char tmp
[1 + 3 * sizeof(int)]; /* < 3 base-10 digits per byte */
280 * PAX attributes have the following layout:
281 * <len> <space> <key> <=> <value> <nl>
283 len
= 1 + (int)strlen(key
) + 1 + (int)strlen(value
) + 1;
286 * The <len> field includes the length of the <len> field, so
287 * computing the correct length is tricky. I start by
288 * counting the number of base-10 digits in 'len' and
289 * computing the next higher power of 10.
297 next_ten
= next_ten
* 10;
300 * For example, if string without the length field is 99
301 * chars, then adding the 2 digit length "99" will force the
302 * total length past 100, requiring an extra digit. The next
303 * statement adjusts for this effect.
305 if (len
+ digits
>= next_ten
)
308 /* Now, we have the right length so we can build the line. */
309 tmp
[sizeof(tmp
) - 1] = 0; /* Null-terminate the work area. */
310 archive_strcat(as
, format_int(tmp
+ sizeof(tmp
) - 1, len
+ digits
));
311 archive_strappend_char(as
, ' ');
312 archive_strcat(as
, key
);
313 archive_strappend_char(as
, '=');
314 archive_strcat(as
, value
);
315 archive_strappend_char(as
, '\n');
319 archive_write_pax_header_xattrs(struct archive_write
*a
,
320 struct pax
*pax
, struct archive_entry
*entry
)
322 struct archive_string s
;
323 int i
= archive_entry_xattr_reset(entry
);
329 char *url_encoded_name
= NULL
, *encoded_name
= NULL
;
333 archive_entry_xattr_next(entry
, &name
, &value
, &size
);
334 url_encoded_name
= url_encode(name
);
335 if (url_encoded_name
!= NULL
) {
336 /* Convert narrow-character to UTF-8. */
337 r
= archive_strcpy_l(&(pax
->l_url_encoded_name
),
338 url_encoded_name
, pax
->sconv_utf8
);
339 free(url_encoded_name
); /* Done with this. */
341 encoded_name
= pax
->l_url_encoded_name
.s
;
342 else if (errno
== ENOMEM
) {
343 archive_set_error(&a
->archive
, ENOMEM
,
344 "Can't allocate memory for Linkname");
345 return (ARCHIVE_FATAL
);
349 encoded_value
= base64_encode((const char *)value
, size
);
351 if (encoded_name
!= NULL
&& encoded_value
!= NULL
) {
352 archive_string_init(&s
);
353 archive_strcpy(&s
, "LIBARCHIVE.xattr.");
354 archive_strcat(&s
, encoded_name
);
355 add_pax_attr(&(pax
->pax_header
), s
.s
, encoded_value
);
356 archive_string_free(&s
);
364 get_entry_hardlink(struct archive_write
*a
, struct archive_entry
*entry
,
365 const char **name
, size_t *length
, struct archive_string_conv
*sc
)
369 r
= archive_entry_hardlink_l(entry
, name
, length
, sc
);
371 if (errno
== ENOMEM
) {
372 archive_set_error(&a
->archive
, ENOMEM
,
373 "Can't allocate memory for Linkname");
374 return (ARCHIVE_FATAL
);
376 return (ARCHIVE_WARN
);
382 get_entry_pathname(struct archive_write
*a
, struct archive_entry
*entry
,
383 const char **name
, size_t *length
, struct archive_string_conv
*sc
)
387 r
= archive_entry_pathname_l(entry
, name
, length
, sc
);
389 if (errno
== ENOMEM
) {
390 archive_set_error(&a
->archive
, ENOMEM
,
391 "Can't allocate memory for Pathname");
392 return (ARCHIVE_FATAL
);
394 return (ARCHIVE_WARN
);
400 get_entry_uname(struct archive_write
*a
, struct archive_entry
*entry
,
401 const char **name
, size_t *length
, struct archive_string_conv
*sc
)
405 r
= archive_entry_uname_l(entry
, name
, length
, sc
);
407 if (errno
== ENOMEM
) {
408 archive_set_error(&a
->archive
, ENOMEM
,
409 "Can't allocate memory for Uname");
410 return (ARCHIVE_FATAL
);
412 return (ARCHIVE_WARN
);
418 get_entry_gname(struct archive_write
*a
, struct archive_entry
*entry
,
419 const char **name
, size_t *length
, struct archive_string_conv
*sc
)
423 r
= archive_entry_gname_l(entry
, name
, length
, sc
);
425 if (errno
== ENOMEM
) {
426 archive_set_error(&a
->archive
, ENOMEM
,
427 "Can't allocate memory for Gname");
428 return (ARCHIVE_FATAL
);
430 return (ARCHIVE_WARN
);
436 get_entry_symlink(struct archive_write
*a
, struct archive_entry
*entry
,
437 const char **name
, size_t *length
, struct archive_string_conv
*sc
)
441 r
= archive_entry_symlink_l(entry
, name
, length
, sc
);
443 if (errno
== ENOMEM
) {
444 archive_set_error(&a
->archive
, ENOMEM
,
445 "Can't allocate memory for Linkname");
446 return (ARCHIVE_FATAL
);
448 return (ARCHIVE_WARN
);
454 * TODO: Consider adding 'comment' and 'charset' fields to
455 * archive_entry so that clients can specify them. Also, consider
456 * adding generic key/value tags so clients can add arbitrary
459 * TODO: Break up this 700-line function!!!! Yowza!
462 archive_write_pax_header(struct archive_write
*a
,
463 struct archive_entry
*entry_original
)
465 struct archive_entry
*entry_main
;
468 int need_extension
, r
, ret
;
470 uint64_t sparse_total
, real_size
;
472 const char *hardlink
;
473 const char *path
= NULL
, *linkpath
= NULL
;
474 const char *uname
= NULL
, *gname
= NULL
;
475 const void *mac_metadata
;
476 size_t mac_metadata_size
;
477 struct archive_string_conv
*sconv
;
478 size_t hardlink_length
, path_length
, linkpath_length
;
479 size_t uname_length
, gname_length
;
483 char ustar_entry_name
[256];
484 char pax_entry_name
[256];
485 char gnu_sparse_name
[256];
486 struct archive_string entry_name
;
490 pax
= (struct pax
*)a
->format_data
;
493 if (archive_entry_pathname(entry_original
) == NULL
) {
494 archive_set_error(&a
->archive
, ARCHIVE_ERRNO_MISC
,
495 "Can't record entry in tar file without pathname");
496 return (ARCHIVE_FAILED
);
500 * Choose a header encoding.
503 sconv
= NULL
;/* Binary mode. */
505 /* Header encoding is UTF-8. */
506 if (pax
->sconv_utf8
== NULL
) {
507 /* Initialize the string conversion object
509 pax
->sconv_utf8
= archive_string_conversion_to_charset(
510 &(a
->archive
), "UTF-8", 1);
511 if (pax
->sconv_utf8
== NULL
)
512 /* Couldn't allocate memory */
513 return (ARCHIVE_FAILED
);
515 sconv
= pax
->sconv_utf8
;
518 r
= get_entry_hardlink(a
, entry_original
, &hardlink
,
519 &hardlink_length
, sconv
);
520 if (r
== ARCHIVE_FATAL
)
522 else if (r
!= ARCHIVE_OK
) {
523 r
= get_entry_hardlink(a
, entry_original
, &hardlink
,
524 &hardlink_length
, NULL
);
525 if (r
== ARCHIVE_FATAL
)
527 archive_set_error(&a
->archive
, ARCHIVE_ERRNO_FILE_FORMAT
,
528 "Can't translate linkname '%s' to %s", hardlink
,
529 archive_string_conversion_charset_name(sconv
));
531 sconv
= NULL
;/* The header charset switches to binary mode. */
534 /* Make sure this is a type of entry that we can handle here */
535 if (hardlink
== NULL
) {
536 switch (archive_entry_filetype(entry_original
)) {
546 * Ensure a trailing '/'. Modify the original
547 * entry so the client sees the change.
549 #if defined(_WIN32) && !defined(__CYGWIN__)
552 wp
= archive_entry_pathname_w(entry_original
);
553 if (wp
!= NULL
&& wp
[wcslen(wp
) -1] != L
'/') {
554 struct archive_wstring ws
;
556 archive_string_init(&ws
);
557 path_length
= wcslen(wp
);
558 if (archive_wstring_ensure(&ws
,
559 path_length
+ 2) == NULL
) {
560 archive_set_error(&a
->archive
, ENOMEM
,
561 "Can't allocate pax data");
562 archive_wstring_free(&ws
);
563 return(ARCHIVE_FATAL
);
565 /* Should we keep '\' ? */
566 if (wp
[path_length
-1] == L
'\\')
568 archive_wstrncpy(&ws
, wp
, path_length
);
569 archive_wstrappend_wchar(&ws
, L
'/');
570 archive_entry_copy_pathname_w(
571 entry_original
, ws
.s
);
572 archive_wstring_free(&ws
);
576 p
= archive_entry_pathname(entry_original
);
578 * On Windows, this is a backup operation just in
579 * case getting WCS failed. On POSIX, this is a
582 if (p
!= NULL
&& p
[strlen(p
) - 1] != '/') {
583 struct archive_string as
;
585 archive_string_init(&as
);
586 path_length
= strlen(p
);
587 if (archive_string_ensure(&as
,
588 path_length
+ 2) == NULL
) {
589 archive_set_error(&a
->archive
, ENOMEM
,
590 "Can't allocate pax data");
591 archive_string_free(&as
);
592 return(ARCHIVE_FATAL
);
594 #if defined(_WIN32) && !defined(__CYGWIN__)
595 /* NOTE: This might break the pathname
596 * if the current code page is CP932 and
597 * the pathname includes a character '\'
598 * as a part of its multibyte pathname. */
599 if (p
[strlen(p
) -1] == '\\')
603 archive_strncpy(&as
, p
, path_length
);
604 archive_strappend_char(&as
, '/');
605 archive_entry_copy_pathname(
606 entry_original
, as
.s
);
607 archive_string_free(&as
);
612 archive_set_error(&a
->archive
,
613 ARCHIVE_ERRNO_FILE_FORMAT
,
614 "tar format cannot archive socket");
615 return (ARCHIVE_FAILED
);
617 archive_set_error(&a
->archive
,
618 ARCHIVE_ERRNO_FILE_FORMAT
,
619 "tar format cannot archive this (type=0%lo)",
621 archive_entry_filetype(entry_original
));
622 return (ARCHIVE_FAILED
);
627 * If Mac OS metadata blob is here, recurse to write that
628 * as a separate entry. This is really a pretty poor design:
629 * In particular, it doubles the overhead for long filenames.
630 * TODO: Help Apple folks design something better and figure
631 * out how to transition from this legacy format.
633 * Note that this code is present on every platform; clients
634 * on non-Mac are unlikely to ever provide this data, but
635 * applications that copy entries from one archive to another
636 * should not lose data just because the local filesystem
640 archive_entry_mac_metadata(entry_original
, &mac_metadata_size
);
641 if (mac_metadata
!= NULL
) {
645 struct archive_entry
*extra
= archive_entry_new2(&a
->archive
);
647 oname
= archive_entry_pathname(entry_original
);
648 name_length
= strlen(oname
);
649 name
= malloc(name_length
+ 3);
650 if (name
== NULL
|| extra
== NULL
) {
651 /* XXX error message */
652 archive_entry_free(extra
);
654 return (ARCHIVE_FAILED
);
657 /* Find last '/'; strip trailing '/' characters */
658 bname
= strrchr(name
, '/');
659 while (bname
!= NULL
&& bname
[1] == '\0') {
661 bname
= strrchr(name
, '/');
664 memmove(name
+ 2, name
, name_length
+ 1);
665 memmove(name
, "._", 2);
668 memmove(bname
+ 2, bname
, strlen(bname
) + 1);
669 memmove(bname
, "._", 2);
671 archive_entry_copy_pathname(extra
, name
);
674 archive_entry_set_size(extra
, mac_metadata_size
);
675 archive_entry_set_filetype(extra
, AE_IFREG
);
676 archive_entry_set_perm(extra
,
677 archive_entry_perm(entry_original
));
678 archive_entry_set_mtime(extra
,
679 archive_entry_mtime(entry_original
),
680 archive_entry_mtime_nsec(entry_original
));
681 archive_entry_set_gid(extra
,
682 archive_entry_gid(entry_original
));
683 archive_entry_set_gname(extra
,
684 archive_entry_gname(entry_original
));
685 archive_entry_set_uid(extra
,
686 archive_entry_uid(entry_original
));
687 archive_entry_set_uname(extra
,
688 archive_entry_uname(entry_original
));
690 /* Recurse to write the special copyfile entry. */
691 r
= archive_write_pax_header(a
, extra
);
692 archive_entry_free(extra
);
693 if (r
< ARCHIVE_WARN
)
697 r
= (int)archive_write_pax_data(a
, mac_metadata
,
699 if (r
< ARCHIVE_WARN
)
703 r
= archive_write_pax_finish_entry(a
);
704 if (r
< ARCHIVE_WARN
)
710 /* Copy entry so we can modify it as needed. */
711 #if defined(_WIN32) && !defined(__CYGWIN__)
712 /* Make sure the path separators in pahtname, hardlink and symlink
713 * are all slash '/', not the Windows path separator '\'. */
714 entry_main
= __la_win_entry_in_posix_pathseparator(entry_original
);
715 if (entry_main
== entry_original
)
716 entry_main
= archive_entry_clone(entry_original
);
718 entry_main
= archive_entry_clone(entry_original
);
720 if (entry_main
== NULL
) {
721 archive_set_error(&a
->archive
, ENOMEM
,
722 "Can't allocate pax data");
723 return(ARCHIVE_FATAL
);
725 archive_string_empty(&(pax
->pax_header
)); /* Blank our work area. */
726 archive_string_empty(&(pax
->sparse_map
));
728 sparse_list_clear(pax
);
730 if (hardlink
== NULL
&&
731 archive_entry_filetype(entry_main
) == AE_IFREG
)
732 sparse_count
= archive_entry_sparse_reset(entry_main
);
736 int64_t offset
, length
, last_offset
= 0;
737 /* Get the last entry of sparse block. */
738 while (archive_entry_sparse_next(
739 entry_main
, &offset
, &length
) == ARCHIVE_OK
)
740 last_offset
= offset
+ length
;
742 /* If the last sparse block does not reach the end of file,
743 * We have to add a empty sparse block as the last entry to
744 * manage storing file data. */
745 if (last_offset
< archive_entry_size(entry_main
))
746 archive_entry_sparse_add_entry(entry_main
,
747 archive_entry_size(entry_main
), 0);
748 sparse_count
= archive_entry_sparse_reset(entry_main
);
752 * First, check the name fields and see if any of them
753 * require binary coding. If any of them does, then all of
756 r
= get_entry_pathname(a
, entry_main
, &path
, &path_length
, sconv
);
757 if (r
== ARCHIVE_FATAL
)
759 else if (r
!= ARCHIVE_OK
) {
760 r
= get_entry_pathname(a
, entry_main
, &path
,
762 if (r
== ARCHIVE_FATAL
)
764 archive_set_error(&a
->archive
, ARCHIVE_ERRNO_FILE_FORMAT
,
765 "Can't translate pathname '%s' to %s", path
,
766 archive_string_conversion_charset_name(sconv
));
768 sconv
= NULL
;/* The header charset switches to binary mode. */
770 r
= get_entry_uname(a
, entry_main
, &uname
, &uname_length
, sconv
);
771 if (r
== ARCHIVE_FATAL
)
773 else if (r
!= ARCHIVE_OK
) {
774 r
= get_entry_uname(a
, entry_main
, &uname
, &uname_length
, NULL
);
775 if (r
== ARCHIVE_FATAL
)
777 archive_set_error(&a
->archive
, ARCHIVE_ERRNO_FILE_FORMAT
,
778 "Can't translate uname '%s' to %s", uname
,
779 archive_string_conversion_charset_name(sconv
));
781 sconv
= NULL
;/* The header charset switches to binary mode. */
783 r
= get_entry_gname(a
, entry_main
, &gname
, &gname_length
, sconv
);
784 if (r
== ARCHIVE_FATAL
)
786 else if (r
!= ARCHIVE_OK
) {
787 r
= get_entry_gname(a
, entry_main
, &gname
, &gname_length
, NULL
);
788 if (r
== ARCHIVE_FATAL
)
790 archive_set_error(&a
->archive
, ARCHIVE_ERRNO_FILE_FORMAT
,
791 "Can't translate gname '%s' to %s", gname
,
792 archive_string_conversion_charset_name(sconv
));
794 sconv
= NULL
;/* The header charset switches to binary mode. */
797 linkpath_length
= hardlink_length
;
798 if (linkpath
== NULL
) {
799 r
= get_entry_symlink(a
, entry_main
, &linkpath
,
800 &linkpath_length
, sconv
);
801 if (r
== ARCHIVE_FATAL
)
803 else if (r
!= ARCHIVE_OK
) {
804 r
= get_entry_symlink(a
, entry_main
, &linkpath
,
805 &linkpath_length
, NULL
);
806 if (r
== ARCHIVE_FATAL
)
808 archive_set_error(&a
->archive
,
809 ARCHIVE_ERRNO_FILE_FORMAT
,
810 "Can't translate linkname '%s' to %s", linkpath
,
811 archive_string_conversion_charset_name(sconv
));
817 /* If any string conversions failed, get all attributes
819 if (sconv
== NULL
&& !pax
->opt_binary
) {
820 if (hardlink
!= NULL
) {
821 r
= get_entry_hardlink(a
, entry_main
, &hardlink
,
822 &hardlink_length
, NULL
);
823 if (r
== ARCHIVE_FATAL
)
826 linkpath_length
= hardlink_length
;
828 r
= get_entry_pathname(a
, entry_main
, &path
,
830 if (r
== ARCHIVE_FATAL
)
832 r
= get_entry_uname(a
, entry_main
, &uname
, &uname_length
, NULL
);
833 if (r
== ARCHIVE_FATAL
)
835 r
= get_entry_gname(a
, entry_main
, &gname
, &gname_length
, NULL
);
836 if (r
== ARCHIVE_FATAL
)
840 /* Store the header encoding first, to be nice to readers. */
842 add_pax_attr(&(pax
->pax_header
), "hdrcharset", "BINARY");
846 * If name is too long, or has non-ASCII characters, add
847 * 'path' to pax extended attrs. (Note that an unconvertible
848 * name must have non-ASCII characters.)
850 if (has_non_ASCII(path
)) {
851 /* We have non-ASCII characters. */
852 add_pax_attr(&(pax
->pax_header
), "path", path
);
853 archive_entry_set_pathname(entry_main
,
854 build_ustar_entry_name(ustar_entry_name
,
855 path
, path_length
, NULL
));
858 /* We have an all-ASCII path; we'd like to just store
859 * it in the ustar header if it will fit. Yes, this
860 * duplicates some of the logic in
861 * archive_write_set_format_ustar.c
863 if (path_length
<= 100) {
864 /* Fits in the old 100-char tar name field. */
866 /* Find largest suffix that will fit. */
867 /* Note: strlen() > 100, so strlen() - 100 - 1 >= 0 */
868 suffix
= strchr(path
+ path_length
- 100 - 1, '/');
869 /* Don't attempt an empty prefix. */
871 suffix
= strchr(suffix
+ 1, '/');
872 /* We can put it in the ustar header if it's
873 * all ASCII and it's either <= 100 characters
874 * or can be split at a '/' into a prefix <=
875 * 155 chars and a suffix <= 100 chars. (Note
876 * the strchr() above will return NULL exactly
877 * when the path can't be split.)
879 if (suffix
== NULL
/* Suffix > 100 chars. */
880 || suffix
[1] == '\0' /* empty suffix */
881 || suffix
- path
> 155) /* Prefix > 155 chars */
883 add_pax_attr(&(pax
->pax_header
), "path", path
);
884 archive_entry_set_pathname(entry_main
,
885 build_ustar_entry_name(ustar_entry_name
,
886 path
, path_length
, NULL
));
892 if (linkpath
!= NULL
) {
893 /* If link name is too long or has non-ASCII characters, add
894 * 'linkpath' to pax extended attrs. */
895 if (linkpath_length
> 100 || has_non_ASCII(linkpath
)) {
896 add_pax_attr(&(pax
->pax_header
), "linkpath", linkpath
);
897 if (linkpath_length
> 100) {
898 if (hardlink
!= NULL
)
899 archive_entry_set_hardlink(entry_main
,
900 "././@LongHardLink");
902 archive_entry_set_symlink(entry_main
,
908 /* Save a pathname since it will be renamed if `entry_main` has
910 archive_string_init(&entry_name
);
911 archive_strcpy(&entry_name
, archive_entry_pathname(entry_main
));
913 /* If file size is too large, add 'size' to pax extended attrs. */
914 if (archive_entry_size(entry_main
) >= (((int64_t)1) << 33)) {
915 add_pax_attr_int(&(pax
->pax_header
), "size",
916 archive_entry_size(entry_main
));
920 /* If numeric GID is too large, add 'gid' to pax extended attrs. */
921 if ((unsigned int)archive_entry_gid(entry_main
) >= (1 << 18)) {
922 add_pax_attr_int(&(pax
->pax_header
), "gid",
923 archive_entry_gid(entry_main
));
927 /* If group name is too large or has non-ASCII characters, add
928 * 'gname' to pax extended attrs. */
930 if (gname_length
> 31 || has_non_ASCII(gname
)) {
931 add_pax_attr(&(pax
->pax_header
), "gname", gname
);
936 /* If numeric UID is too large, add 'uid' to pax extended attrs. */
937 if ((unsigned int)archive_entry_uid(entry_main
) >= (1 << 18)) {
938 add_pax_attr_int(&(pax
->pax_header
), "uid",
939 archive_entry_uid(entry_main
));
943 /* Add 'uname' to pax extended attrs if necessary. */
945 if (uname_length
> 31 || has_non_ASCII(uname
)) {
946 add_pax_attr(&(pax
->pax_header
), "uname", uname
);
952 * POSIX/SUSv3 doesn't provide a standard key for large device
953 * numbers. I use the same keys here that Joerg Schilling
954 * used for 'star.' (Which, somewhat confusingly, are called
955 * "devXXX" even though they code "rdev" values.) No doubt,
956 * other implementations use other keys. Note that there's no
957 * reason we can't write the same information into a number of
960 * Of course, this is only needed for block or char device entries.
962 if (archive_entry_filetype(entry_main
) == AE_IFBLK
963 || archive_entry_filetype(entry_main
) == AE_IFCHR
) {
965 * If rdevmajor is too large, add 'SCHILY.devmajor' to
966 * extended attributes.
968 int rdevmajor
, rdevminor
;
969 rdevmajor
= archive_entry_rdevmajor(entry_main
);
970 rdevminor
= archive_entry_rdevminor(entry_main
);
971 if (rdevmajor
>= (1 << 18)) {
972 add_pax_attr_int(&(pax
->pax_header
), "SCHILY.devmajor",
975 * Non-strict formatting below means we don't
976 * have to truncate here. Not truncating improves
977 * the chance that some more modern tar archivers
978 * (such as GNU tar 1.13) can restore the full
979 * value even if they don't understand the pax
980 * extended attributes. See my rant below about
981 * file size fields for additional details.
983 /* archive_entry_set_rdevmajor(entry_main,
984 rdevmajor & ((1 << 18) - 1)); */
989 * If devminor is too large, add 'SCHILY.devminor' to
990 * extended attributes.
992 if (rdevminor
>= (1 << 18)) {
993 add_pax_attr_int(&(pax
->pax_header
), "SCHILY.devminor",
995 /* Truncation is not necessary here, either. */
996 /* archive_entry_set_rdevminor(entry_main,
997 rdevminor & ((1 << 18) - 1)); */
1003 * Technically, the mtime field in the ustar header can
1004 * support 33 bits, but many platforms use signed 32-bit time
1005 * values. The cutoff of 0x7fffffff here is a compromise.
1006 * Yes, this check is duplicated just below; this helps to
1007 * avoid writing an mtime attribute just to handle a
1008 * high-resolution timestamp in "restricted pax" mode.
1010 if (!need_extension
&&
1011 ((archive_entry_mtime(entry_main
) < 0)
1012 || (archive_entry_mtime(entry_main
) >= 0x7fffffff)))
1015 /* I use a star-compatible file flag attribute. */
1016 p
= archive_entry_fflags_text(entry_main
);
1017 if (!need_extension
&& p
!= NULL
&& *p
!= '\0')
1020 /* If there are non-trivial ACL entries, we need an extension. */
1021 if (!need_extension
&& archive_entry_acl_count(entry_original
,
1022 ARCHIVE_ENTRY_ACL_TYPE_ACCESS
) > 0)
1025 /* If there are non-trivial ACL entries, we need an extension. */
1026 if (!need_extension
&& archive_entry_acl_count(entry_original
,
1027 ARCHIVE_ENTRY_ACL_TYPE_DEFAULT
) > 0)
1030 /* If there are extended attributes, we need an extension */
1031 if (!need_extension
&& archive_entry_xattr_count(entry_original
) > 0)
1034 /* If there are sparse info, we need an extension */
1035 if (!need_extension
&& sparse_count
> 0)
1039 * Libarchive used to include these in extended headers for
1040 * restricted pax format, but that confused people who
1041 * expected ustar-like time semantics. So now we only include
1042 * them in full pax format.
1044 if (a
->archive
.archive_format
!= ARCHIVE_FORMAT_TAR_PAX_RESTRICTED
) {
1045 if (archive_entry_ctime(entry_main
) != 0 ||
1046 archive_entry_ctime_nsec(entry_main
) != 0)
1047 add_pax_attr_time(&(pax
->pax_header
), "ctime",
1048 archive_entry_ctime(entry_main
),
1049 archive_entry_ctime_nsec(entry_main
));
1051 if (archive_entry_atime(entry_main
) != 0 ||
1052 archive_entry_atime_nsec(entry_main
) != 0)
1053 add_pax_attr_time(&(pax
->pax_header
), "atime",
1054 archive_entry_atime(entry_main
),
1055 archive_entry_atime_nsec(entry_main
));
1057 /* Store birth/creationtime only if it's earlier than mtime */
1058 if (archive_entry_birthtime_is_set(entry_main
) &&
1059 archive_entry_birthtime(entry_main
)
1060 < archive_entry_mtime(entry_main
))
1061 add_pax_attr_time(&(pax
->pax_header
),
1062 "LIBARCHIVE.creationtime",
1063 archive_entry_birthtime(entry_main
),
1064 archive_entry_birthtime_nsec(entry_main
));
1068 * The following items are handled differently in "pax
1069 * restricted" format. In particular, in "pax restricted"
1070 * format they won't be added unless need_extension is
1071 * already set (we're already generating an extended header, so
1072 * may as well include these).
1074 if (a
->archive
.archive_format
!= ARCHIVE_FORMAT_TAR_PAX_RESTRICTED
||
1076 if (archive_entry_mtime(entry_main
) < 0 ||
1077 archive_entry_mtime(entry_main
) >= 0x7fffffff ||
1078 archive_entry_mtime_nsec(entry_main
) != 0)
1079 add_pax_attr_time(&(pax
->pax_header
), "mtime",
1080 archive_entry_mtime(entry_main
),
1081 archive_entry_mtime_nsec(entry_main
));
1083 /* I use a star-compatible file flag attribute. */
1084 p
= archive_entry_fflags_text(entry_main
);
1085 if (p
!= NULL
&& *p
!= '\0')
1086 add_pax_attr(&(pax
->pax_header
), "SCHILY.fflags", p
);
1088 /* I use star-compatible ACL attributes. */
1089 r
= archive_entry_acl_text_l(entry_original
,
1090 ARCHIVE_ENTRY_ACL_TYPE_ACCESS
|
1091 ARCHIVE_ENTRY_ACL_STYLE_EXTRA_ID
,
1092 &p
, NULL
, pax
->sconv_utf8
);
1094 if (errno
== ENOMEM
) {
1095 archive_set_error(&a
->archive
, ENOMEM
,
1096 "Can't allocate memory for "
1098 return (ARCHIVE_FATAL
);
1100 archive_set_error(&a
->archive
,
1101 ARCHIVE_ERRNO_FILE_FORMAT
,
1102 "Can't translate ACL.access to UTF-8");
1104 } else if (p
!= NULL
&& *p
!= '\0') {
1105 add_pax_attr(&(pax
->pax_header
),
1106 "SCHILY.acl.access", p
);
1108 r
= archive_entry_acl_text_l(entry_original
,
1109 ARCHIVE_ENTRY_ACL_TYPE_DEFAULT
|
1110 ARCHIVE_ENTRY_ACL_STYLE_EXTRA_ID
,
1111 &p
, NULL
, pax
->sconv_utf8
);
1113 if (errno
== ENOMEM
) {
1114 archive_set_error(&a
->archive
, ENOMEM
,
1115 "Can't allocate memory for "
1117 return (ARCHIVE_FATAL
);
1119 archive_set_error(&a
->archive
,
1120 ARCHIVE_ERRNO_FILE_FORMAT
,
1121 "Can't translate ACL.default to UTF-8");
1123 } else if (p
!= NULL
&& *p
!= '\0') {
1124 add_pax_attr(&(pax
->pax_header
),
1125 "SCHILY.acl.default", p
);
1128 /* We use GNU-tar-compatible sparse attributes. */
1129 if (sparse_count
> 0) {
1130 int64_t soffset
, slength
;
1132 add_pax_attr_int(&(pax
->pax_header
),
1133 "GNU.sparse.major", 1);
1134 add_pax_attr_int(&(pax
->pax_header
),
1135 "GNU.sparse.minor", 0);
1136 add_pax_attr(&(pax
->pax_header
),
1137 "GNU.sparse.name", entry_name
.s
);
1138 add_pax_attr_int(&(pax
->pax_header
),
1139 "GNU.sparse.realsize",
1140 archive_entry_size(entry_main
));
1142 /* Rename the file name which will be used for
1143 * ustar header to a special name, which GNU
1144 * PAX Format 1.0 requires */
1145 archive_entry_set_pathname(entry_main
,
1146 build_gnu_sparse_name(gnu_sparse_name
,
1150 * - Make a sparse map, which will precede a file data.
1151 * - Get the total size of available data of sparse.
1153 archive_string_sprintf(&(pax
->sparse_map
), "%d\n",
1155 while (archive_entry_sparse_next(entry_main
,
1156 &soffset
, &slength
) == ARCHIVE_OK
) {
1157 archive_string_sprintf(&(pax
->sparse_map
),
1161 sparse_total
+= slength
;
1162 if (sparse_list_add(pax
, soffset
, slength
)
1164 archive_set_error(&a
->archive
,
1166 "Can't allocate memory");
1167 archive_entry_free(entry_main
);
1168 archive_string_free(&entry_name
);
1169 return (ARCHIVE_FATAL
);
1174 /* Store extended attributes */
1175 if (archive_write_pax_header_xattrs(a
, pax
, entry_original
)
1177 archive_entry_free(entry_main
);
1178 archive_string_free(&entry_name
);
1179 return (ARCHIVE_FATAL
);
1183 /* Only regular files have data. */
1184 if (archive_entry_filetype(entry_main
) != AE_IFREG
)
1185 archive_entry_set_size(entry_main
, 0);
1188 * Pax-restricted does not store data for hardlinks, in order
1189 * to improve compatibility with ustar.
1191 if (a
->archive
.archive_format
!= ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE
&&
1193 archive_entry_set_size(entry_main
, 0);
1196 * XXX Full pax interchange format does permit a hardlink
1197 * entry to have data associated with it. I'm not supporting
1198 * that here because the client expects me to tell them whether
1199 * or not this format expects data for hardlinks. If I
1200 * don't check here, then every pax archive will end up with
1201 * duplicated data for hardlinks. Someday, there may be
1202 * need to select this behavior, in which case the following
1203 * will need to be revisited. XXX
1205 if (hardlink
!= NULL
)
1206 archive_entry_set_size(entry_main
, 0);
1208 /* Save a real file size. */
1209 real_size
= archive_entry_size(entry_main
);
1211 * Overwrite a file size by the total size of sparse blocks and
1212 * the size of sparse map info. That file size is the length of
1213 * the data, which we will exactly store into an archive file.
1215 if (archive_strlen(&(pax
->sparse_map
))) {
1216 size_t mapsize
= archive_strlen(&(pax
->sparse_map
));
1217 pax
->sparse_map_padding
= 0x1ff & (-(ssize_t
)mapsize
);
1218 archive_entry_set_size(entry_main
,
1219 mapsize
+ pax
->sparse_map_padding
+ sparse_total
);
1222 /* Format 'ustar' header for main entry.
1224 * The trouble with file size: If the reader can't understand
1225 * the file size, they may not be able to locate the next
1226 * entry and the rest of the archive is toast. Pax-compliant
1227 * readers are supposed to ignore the file size in the main
1228 * header, so the question becomes how to maximize portability
1229 * for readers that don't support pax attribute extensions.
1230 * For maximum compatibility, I permit numeric extensions in
1231 * the main header so that the file size stored will always be
1232 * correct, even if it's in a format that only some
1233 * implementations understand. The technique used here is:
1235 * a) If possible, follow the standard exactly. This handles
1236 * files up to 8 gigabytes minus 1.
1238 * b) If that fails, try octal but omit the field terminator.
1239 * That handles files up to 64 gigabytes minus 1.
1241 * c) Otherwise, use base-256 extensions. That handles files
1242 * up to 2^63 in this implementation, with the potential to
1243 * go up to 2^94. That should hold us for a while. ;-)
1245 * The non-strict formatter uses similar logic for other
1246 * numeric fields, though they're less critical.
1248 if (__archive_write_format_header_ustar(a
, ustarbuff
, entry_main
, -1, 0,
1249 NULL
) == ARCHIVE_FATAL
)
1250 return (ARCHIVE_FATAL
);
1252 /* If we built any extended attributes, write that entry first. */
1253 if (archive_strlen(&(pax
->pax_header
)) > 0) {
1254 struct archive_entry
*pax_attr_entry
;
1259 pax_attr_entry
= archive_entry_new2(&a
->archive
);
1261 archive_entry_set_pathname(pax_attr_entry
,
1262 build_pax_attribute_name(pax_entry_name
, p
));
1263 archive_entry_set_size(pax_attr_entry
,
1264 archive_strlen(&(pax
->pax_header
)));
1265 /* Copy uid/gid (but clip to ustar limits). */
1266 uid
= archive_entry_uid(entry_main
);
1268 uid
= (1 << 18) - 1;
1269 archive_entry_set_uid(pax_attr_entry
, uid
);
1270 gid
= archive_entry_gid(entry_main
);
1272 gid
= (1 << 18) - 1;
1273 archive_entry_set_gid(pax_attr_entry
, gid
);
1274 /* Copy mode over (but not setuid/setgid bits) */
1275 mode
= archive_entry_mode(entry_main
);
1285 archive_entry_set_mode(pax_attr_entry
, mode
);
1287 /* Copy uname/gname. */
1288 archive_entry_set_uname(pax_attr_entry
,
1289 archive_entry_uname(entry_main
));
1290 archive_entry_set_gname(pax_attr_entry
,
1291 archive_entry_gname(entry_main
));
1293 /* Copy mtime, but clip to ustar limits. */
1294 s
= archive_entry_mtime(entry_main
);
1295 if (s
< 0) { s
= 0; }
1296 if (s
>= 0x7fffffff) { s
= 0x7fffffff; }
1297 archive_entry_set_mtime(pax_attr_entry
, s
, 0);
1299 /* Standard ustar doesn't support atime. */
1300 archive_entry_set_atime(pax_attr_entry
, 0, 0);
1302 /* Standard ustar doesn't support ctime. */
1303 archive_entry_set_ctime(pax_attr_entry
, 0, 0);
1305 r
= __archive_write_format_header_ustar(a
, paxbuff
,
1306 pax_attr_entry
, 'x', 1, NULL
);
1308 archive_entry_free(pax_attr_entry
);
1310 /* Note that the 'x' header shouldn't ever fail to format */
1311 if (r
< ARCHIVE_WARN
) {
1312 archive_set_error(&a
->archive
, ARCHIVE_ERRNO_MISC
,
1313 "archive_write_pax_header: "
1314 "'x' header failed?! This can't happen.\n");
1315 return (ARCHIVE_FATAL
);
1318 r
= __archive_write_output(a
, paxbuff
, 512);
1319 if (r
!= ARCHIVE_OK
) {
1320 sparse_list_clear(pax
);
1321 pax
->entry_bytes_remaining
= 0;
1322 pax
->entry_padding
= 0;
1323 return (ARCHIVE_FATAL
);
1326 pax
->entry_bytes_remaining
= archive_strlen(&(pax
->pax_header
));
1327 pax
->entry_padding
=
1328 0x1ff & (-(int64_t)pax
->entry_bytes_remaining
);
1330 r
= __archive_write_output(a
, pax
->pax_header
.s
,
1331 archive_strlen(&(pax
->pax_header
)));
1332 if (r
!= ARCHIVE_OK
) {
1333 /* If a write fails, we're pretty much toast. */
1334 return (ARCHIVE_FATAL
);
1336 /* Pad out the end of the entry. */
1337 r
= __archive_write_nulls(a
, (size_t)pax
->entry_padding
);
1338 if (r
!= ARCHIVE_OK
) {
1339 /* If a write fails, we're pretty much toast. */
1340 return (ARCHIVE_FATAL
);
1342 pax
->entry_bytes_remaining
= pax
->entry_padding
= 0;
1345 /* Write the header for main entry. */
1346 r
= __archive_write_output(a
, ustarbuff
, 512);
1347 if (r
!= ARCHIVE_OK
)
1351 * Inform the client of the on-disk size we're using, so
1352 * they can avoid unnecessarily writing a body for something
1353 * that we're just going to ignore.
1355 archive_entry_set_size(entry_original
, real_size
);
1356 if (pax
->sparse_list
== NULL
&& real_size
> 0) {
1357 /* This is not a sparse file but we handle its data as
1358 * a sparse block. */
1359 sparse_list_add(pax
, 0, real_size
);
1360 sparse_total
= real_size
;
1362 pax
->entry_padding
= 0x1ff & (-(int64_t)sparse_total
);
1363 archive_entry_free(entry_main
);
1364 archive_string_free(&entry_name
);
1370 * We need a valid name for the regular 'ustar' entry. This routine
1371 * tries to hack something more-or-less reasonable.
1373 * The approach here tries to preserve leading dir names. We do so by
1374 * working with four sections:
1375 * 1) "prefix" directory names,
1376 * 2) "suffix" directory names,
1377 * 3) inserted dir name (optional),
1380 * These sections must satisfy the following requirements:
1381 * * Parts 1 & 2 together form an initial portion of the dir name.
1382 * * Part 3 is specified by the caller. (It should not contain a leading
1384 * * Part 4 forms an initial portion of the base filename.
1385 * * The filename must be <= 99 chars to fit the ustar 'name' field.
1386 * * Parts 2, 3, 4 together must be <= 99 chars to fit the ustar 'name' fld.
1387 * * Part 1 must be <= 155 chars to fit the ustar 'prefix' field.
1388 * * If the original name ends in a '/', the new name must also end in a '/'
1389 * * Trailing '/.' sequences may be stripped.
1391 * Note: Recall that the ustar format does not store the '/' separating
1392 * parts 1 & 2, but does store the '/' separating parts 2 & 3.
1395 build_ustar_entry_name(char *dest
, const char *src
, size_t src_length
,
1398 const char *prefix
, *prefix_end
;
1399 const char *suffix
, *suffix_end
;
1400 const char *filename
, *filename_end
;
1402 int need_slash
= 0; /* Was there a trailing slash? */
1403 size_t suffix_length
= 99;
1404 size_t insert_length
;
1406 /* Length of additional dir element to be added. */
1410 /* +2 here allows for '/' before and after the insert. */
1411 insert_length
= strlen(insert
) + 2;
1413 /* Step 0: Quick bailout in a common case. */
1414 if (src_length
< 100 && insert
== NULL
) {
1415 strncpy(dest
, src
, src_length
);
1416 dest
[src_length
] = '\0';
1420 /* Step 1: Locate filename and enforce the length restriction. */
1421 filename_end
= src
+ src_length
;
1422 /* Remove trailing '/' chars and '/.' pairs. */
1424 if (filename_end
> src
&& filename_end
[-1] == '/') {
1426 need_slash
= 1; /* Remember to restore trailing '/'. */
1429 if (filename_end
> src
+ 1 && filename_end
[-1] == '.'
1430 && filename_end
[-2] == '/') {
1432 need_slash
= 1; /* "foo/." will become "foo/" */
1439 /* Find start of filename. */
1440 filename
= filename_end
- 1;
1441 while ((filename
> src
) && (*filename
!= '/'))
1443 if ((*filename
== '/') && (filename
< filename_end
- 1))
1445 /* Adjust filename_end so that filename + insert fits in 99 chars. */
1446 suffix_length
-= insert_length
;
1447 if (filename_end
> filename
+ suffix_length
)
1448 filename_end
= filename
+ suffix_length
;
1449 /* Calculate max size for "suffix" section (#3 above). */
1450 suffix_length
-= filename_end
- filename
;
1452 /* Step 2: Locate the "prefix" section of the dirname, including
1455 prefix_end
= prefix
+ 155;
1456 if (prefix_end
> filename
)
1457 prefix_end
= filename
;
1458 while (prefix_end
> prefix
&& *prefix_end
!= '/')
1460 if ((prefix_end
< filename
) && (*prefix_end
== '/'))
1463 /* Step 3: Locate the "suffix" section of the dirname,
1464 * including trailing '/'. */
1465 suffix
= prefix_end
;
1466 suffix_end
= suffix
+ suffix_length
; /* Enforce limit. */
1467 if (suffix_end
> filename
)
1468 suffix_end
= filename
;
1469 if (suffix_end
< suffix
)
1470 suffix_end
= suffix
;
1471 while (suffix_end
> suffix
&& *suffix_end
!= '/')
1473 if ((suffix_end
< filename
) && (*suffix_end
== '/'))
1476 /* Step 4: Build the new name. */
1477 /* The OpenBSD strlcpy function is safer, but less portable. */
1478 /* Rather than maintain two versions, just use the strncpy version. */
1480 if (prefix_end
> prefix
) {
1481 strncpy(p
, prefix
, prefix_end
- prefix
);
1482 p
+= prefix_end
- prefix
;
1484 if (suffix_end
> suffix
) {
1485 strncpy(p
, suffix
, suffix_end
- suffix
);
1486 p
+= suffix_end
- suffix
;
1488 if (insert
!= NULL
) {
1489 /* Note: assume insert does not have leading or trailing '/' */
1491 p
+= strlen(insert
);
1494 strncpy(p
, filename
, filename_end
- filename
);
1495 p
+= filename_end
- filename
;
1504 * The ustar header for the pax extended attributes must have a
1505 * reasonable name: SUSv3 requires 'dirname'/PaxHeader.'pid'/'filename'
1506 * where 'pid' is the PID of the archiving process. Unfortunately,
1507 * that makes testing a pain since the output varies for each run,
1508 * so I'm sticking with the simpler 'dirname'/PaxHeader/'filename'
1509 * for now. (Someday, I'll make this settable. Then I can use the
1510 * SUS recommendation as default and test harnesses can override it
1511 * to get predictable results.)
1513 * Joerg Schilling has argued that this is unnecessary because, in
1514 * practice, if the pax extended attributes get extracted as regular
1515 * files, no one is going to bother reading those attributes to
1516 * manually restore them. Based on this, 'star' uses
1517 * /tmp/PaxHeader/'basename' as the ustar header name. This is a
1518 * tempting argument, in part because it's simpler than the SUSv3
1519 * recommendation, but I'm not entirely convinced. I'm also
1520 * uncomfortable with the fact that "/tmp" is a Unix-ism.
1522 * The following routine leverages build_ustar_entry_name() above and
1523 * so is simpler than you might think. It just needs to provide the
1524 * additional path element and handle a few pathological cases).
1527 build_pax_attribute_name(char *dest
, const char *src
)
1532 /* Handle the null filename case. */
1533 if (src
== NULL
|| *src
== '\0') {
1534 strcpy(dest
, "PaxHeader/blank");
1538 /* Prune final '/' and other unwanted final elements. */
1539 p
= src
+ strlen(src
);
1541 /* Ends in "/", remove the '/' */
1542 if (p
> src
&& p
[-1] == '/') {
1546 /* Ends in "/.", remove the '.' */
1547 if (p
> src
+ 1 && p
[-1] == '.'
1555 /* Pathological case: After above, there was nothing left.
1556 * This includes "/." "/./." "/.//./." etc. */
1558 strcpy(dest
, "/PaxHeader/rootdir");
1562 /* Convert unadorned "." into a suitable filename. */
1563 if (*src
== '.' && p
== src
+ 1) {
1564 strcpy(dest
, "PaxHeader/currentdir");
1569 * TODO: Push this string into the 'pax' structure to avoid
1570 * recomputing it every time. That will also open the door
1571 * to having clients override it.
1573 #if HAVE_GETPID && 0 /* Disable this for now; see above comment. */
1574 sprintf(buff
, "PaxHeader.%d", getpid());
1576 /* If the platform can't fetch the pid, don't include it. */
1577 strcpy(buff
, "PaxHeader");
1579 /* General case: build a ustar-compatible name adding
1581 build_ustar_entry_name(dest
, src
, p
- src
, buff
);
1587 * GNU PAX Format 1.0 requires the special name, which pattern is:
1588 * <dir>/GNUSparseFile.<pid>/<original file name>
1590 * This function is used for only Sparse file, a file type of which
1594 build_gnu_sparse_name(char *dest
, const char *src
)
1599 /* Handle the null filename case. */
1600 if (src
== NULL
|| *src
== '\0') {
1601 strcpy(dest
, "GNUSparseFile/blank");
1605 /* Prune final '/' and other unwanted final elements. */
1606 p
= src
+ strlen(src
);
1608 /* Ends in "/", remove the '/' */
1609 if (p
> src
&& p
[-1] == '/') {
1613 /* Ends in "/.", remove the '.' */
1614 if (p
> src
+ 1 && p
[-1] == '.'
1622 #if HAVE_GETPID && 0 /* Disable this as pax attribute name. */
1623 sprintf(buff
, "GNUSparseFile.%d", getpid());
1625 /* If the platform can't fetch the pid, don't include it. */
1626 strcpy(buff
, "GNUSparseFile");
1628 /* General case: build a ustar-compatible name adding
1629 * "/GNUSparseFile/". */
1630 build_ustar_entry_name(dest
, src
, p
- src
, buff
);
1635 /* Write two null blocks for the end of archive */
1637 archive_write_pax_close(struct archive_write
*a
)
1639 return (__archive_write_nulls(a
, 512 * 2));
1643 archive_write_pax_free(struct archive_write
*a
)
1647 pax
= (struct pax
*)a
->format_data
;
1649 return (ARCHIVE_OK
);
1651 archive_string_free(&pax
->pax_header
);
1652 archive_string_free(&pax
->sparse_map
);
1653 archive_string_free(&pax
->l_url_encoded_name
);
1654 sparse_list_clear(pax
);
1656 a
->format_data
= NULL
;
1657 return (ARCHIVE_OK
);
1661 archive_write_pax_finish_entry(struct archive_write
*a
)
1667 pax
= (struct pax
*)a
->format_data
;
1668 remaining
= pax
->entry_bytes_remaining
;
1669 if (remaining
== 0) {
1670 while (pax
->sparse_list
) {
1671 struct sparse_block
*sb
;
1672 if (!pax
->sparse_list
->is_hole
)
1673 remaining
+= pax
->sparse_list
->remaining
;
1674 sb
= pax
->sparse_list
->next
;
1675 free(pax
->sparse_list
);
1676 pax
->sparse_list
= sb
;
1679 ret
= __archive_write_nulls(a
, (size_t)(remaining
+ pax
->entry_padding
));
1680 pax
->entry_bytes_remaining
= pax
->entry_padding
= 0;
1685 archive_write_pax_data(struct archive_write
*a
, const void *buff
, size_t s
)
1692 pax
= (struct pax
*)a
->format_data
;
1695 * According to GNU PAX format 1.0, write a sparse map
1698 if (archive_strlen(&(pax
->sparse_map
))) {
1699 ret
= __archive_write_output(a
, pax
->sparse_map
.s
,
1700 archive_strlen(&(pax
->sparse_map
)));
1701 if (ret
!= ARCHIVE_OK
)
1703 ret
= __archive_write_nulls(a
, pax
->sparse_map_padding
);
1704 if (ret
!= ARCHIVE_OK
)
1706 archive_string_empty(&(pax
->sparse_map
));
1711 const unsigned char *p
;
1713 while (pax
->sparse_list
!= NULL
&&
1714 pax
->sparse_list
->remaining
== 0) {
1715 struct sparse_block
*sb
= pax
->sparse_list
->next
;
1716 free(pax
->sparse_list
);
1717 pax
->sparse_list
= sb
;
1720 if (pax
->sparse_list
== NULL
)
1723 p
= ((const unsigned char *)buff
) + total
;
1725 if (ws
> pax
->sparse_list
->remaining
)
1726 ws
= (size_t)pax
->sparse_list
->remaining
;
1728 if (pax
->sparse_list
->is_hole
) {
1729 /* Current block is hole thus we do not write
1731 pax
->sparse_list
->remaining
-= ws
;
1736 ret
= __archive_write_output(a
, p
, ws
);
1737 pax
->sparse_list
->remaining
-= ws
;
1739 if (ret
!= ARCHIVE_OK
)
1746 has_non_ASCII(const char *_p
)
1748 const unsigned char *p
= (const unsigned char *)_p
;
1752 while (*p
!= '\0' && *p
< 128)
1754 return (*p
!= '\0');
1758 * Used by extended attribute support; encodes the name
1759 * so that there will be no '=' characters in the result.
1762 url_encode(const char *in
)
1769 for (s
= in
; *s
!= '\0'; s
++) {
1770 if (*s
< 33 || *s
> 126 || *s
== '%' || *s
== '=')
1776 out
= (char *)malloc(out_len
+ 1);
1780 for (s
= in
, d
= out
; *s
!= '\0'; s
++) {
1781 /* encode any non-printable ASCII character or '%' or '=' */
1782 if (*s
< 33 || *s
> 126 || *s
== '%' || *s
== '=') {
1783 /* URL encoding is '%' followed by two hex digits */
1785 *d
++ = "0123456789ABCDEF"[0x0f & (*s
>> 4)];
1786 *d
++ = "0123456789ABCDEF"[0x0f & *s
];
1796 * Encode a sequence of bytes into a C string using base-64 encoding.
1798 * Returns a null-terminated C string allocated with malloc(); caller
1799 * is responsible for freeing the result.
1802 base64_encode(const char *s
, size_t len
)
1804 static const char digits
[64] =
1805 { 'A','B','C','D','E','F','G','H','I','J','K','L','M','N','O',
1806 'P','Q','R','S','T','U','V','W','X','Y','Z','a','b','c','d',
1807 'e','f','g','h','i','j','k','l','m','n','o','p','q','r','s',
1808 't','u','v','w','x','y','z','0','1','2','3','4','5','6','7',
1813 /* 3 bytes becomes 4 chars, but round up and allow for trailing NUL */
1814 out
= (char *)malloc((len
* 4 + 2) / 3 + 1);
1819 /* Convert each group of 3 bytes into 4 characters. */
1821 v
= (((int)s
[0] << 16) & 0xff0000)
1822 | (((int)s
[1] << 8) & 0xff00)
1823 | (((int)s
[2]) & 0x00ff);
1826 *d
++ = digits
[(v
>> 18) & 0x3f];
1827 *d
++ = digits
[(v
>> 12) & 0x3f];
1828 *d
++ = digits
[(v
>> 6) & 0x3f];
1829 *d
++ = digits
[(v
) & 0x3f];
1831 /* Handle final group of 1 byte (2 chars) or 2 bytes (3 chars). */
1835 v
= (((int)s
[0] << 16) & 0xff0000);
1836 *d
++ = digits
[(v
>> 18) & 0x3f];
1837 *d
++ = digits
[(v
>> 12) & 0x3f];
1840 v
= (((int)s
[0] << 16) & 0xff0000)
1841 | (((int)s
[1] << 8) & 0xff00);
1842 *d
++ = digits
[(v
>> 18) & 0x3f];
1843 *d
++ = digits
[(v
>> 12) & 0x3f];
1844 *d
++ = digits
[(v
>> 6) & 0x3f];
1847 /* Add trailing NUL character so output is a valid C string. */
1853 sparse_list_clear(struct pax
*pax
)
1855 while (pax
->sparse_list
!= NULL
) {
1856 struct sparse_block
*sb
= pax
->sparse_list
;
1857 pax
->sparse_list
= sb
->next
;
1860 pax
->sparse_tail
= NULL
;
1864 _sparse_list_add_block(struct pax
*pax
, int64_t offset
, int64_t length
,
1867 struct sparse_block
*sb
;
1869 sb
= (struct sparse_block
*)malloc(sizeof(*sb
));
1871 return (ARCHIVE_FATAL
);
1873 sb
->is_hole
= is_hole
;
1874 sb
->offset
= offset
;
1875 sb
->remaining
= length
;
1876 if (pax
->sparse_list
== NULL
|| pax
->sparse_tail
== NULL
)
1877 pax
->sparse_list
= pax
->sparse_tail
= sb
;
1879 pax
->sparse_tail
->next
= sb
;
1880 pax
->sparse_tail
= sb
;
1882 return (ARCHIVE_OK
);
1886 sparse_list_add(struct pax
*pax
, int64_t offset
, int64_t length
)
1888 int64_t last_offset
;
1891 if (pax
->sparse_tail
== NULL
)
1894 last_offset
= pax
->sparse_tail
->offset
+
1895 pax
->sparse_tail
->remaining
;
1897 if (last_offset
< offset
) {
1898 /* Add a hole block. */
1899 r
= _sparse_list_add_block(pax
, last_offset
,
1900 offset
- last_offset
, 1);
1901 if (r
!= ARCHIVE_OK
)
1904 /* Add data block. */
1905 return (_sparse_list_add_block(pax
, offset
, length
, 0));